Himalaya Dua’s 2nd place NVIDIA hackathon win with a last-minute team
Hours after the Wi-Fi went down at the Hyatt Regency San Francisco Airport, Himalaya Dua sat amid hundreds of animated voices at the only empty table in the hotel conference hall. Although his team of four other students had waited to see if the internet would be restored, they eventually had to leave, and Himalaya was on his own at the NVIDIA Hackathon at ODSC West 2024.
They had arrived early, eager to network with other AI enthusiasts and get hands-on experience with NVIDIA’s cutting-edge tools at one of the world’s leading AI conferences. But when a record-breaking number of participants downloaded the hackathon’s massive dataset at once, the network collapsed, throwing an unexpected hurdle their way.
For a while, everyone made do with mobile hotspots. But as the hours passed, it became unclear whether the competition could continue. When one of Himalaya’s teammates—the one who had driven them there—finally needed to leave, the others, not wanting to be stranded an hour from home, followed.
Except for Himalaya. He hated the thought of giving up.
Years earlier, when he was captain of his school’s basketball team, he had made the call to walk away from something—and he still regretted it. So when the others got up to leave, he couldn’t. He thought about that moment. He thought about the importance of sitting with uncertainty and keeping the promises you make to yourself. “When you don’t keep your promises, your brain learns that walking away is an option,” he said. “And then walking away becomes easier over time.”
So he stayed—uncertain, but determined to keep going.
Ten minutes later, a woman approached him. Then a man asked to join them. She was Sara Zare, an AI investor and former Google product leader. He was Feifan Liu, an associate professor at UMass Chan Medical School. Like Himalaya, they had never competed in a hackathon before. Yet their impromptu team finished among the top three, winning first place in accuracy and second place overall.
Their unexpected win reinforced what Himalaya already believed: the right people appear when you need them—if you put yourself in their path.
An openness to new people and new paths
Himalaya’s first influences were his parents. Relentless and hardworking—he said he never once saw them sit down to watch TV—they taught him to make the most of his time and never run from a challenge.
He took that to heart. In school, he sought out the toughest courses, just to “see what everyone was afraid of.” In sports, he was competitive. And when it came time to choose his undergraduate major, he went for information technology—because it was harder.
Later, in his tech career, mentors pushed him to go beyond just doing his job and leave his mark. Himalaya never sought them out—just as he didn’t seek out the two strangers who became his hackathon teammates. But, he said, the right people always seem to find him at the right time. “I’ve always been very, very lucky.”
His roommates are further proof. He hadn’t planned to share an apartment with other Computer Science master’s students who are as driven as he is—but that’s what happened. After connecting online, they bonded over their shared commitment to professional growth, exchanging LinkedIn profiles instead of Instagram or Facebook. “We sit down and talk about tech. We work on things. They push me forward,” he said.
One of his roommates registered him for the NVIDIA Hackathon without his knowledge, then casually informed him later. “He knew I wouldn’t say no.” That roommate later realized he had an exam and dropped out, forcing Himalaya to search for teammates in a student WhatsApp group and on Viva Engage. Then at the Hyatt hotel he ended up alone again—until the right people found him.
Himalaya attending a conference at Google
Meeting the challenge with adaptability
With over 1,000 participants and more than 200 teams competing in person, the NVIDIA Hackathon in San Francisco was ODSC’s biggest yet. Open to all skill levels, it attracted students, researchers, and tech professionals—including NVIDIA employees.
A 24-hour time limit, advanced technical demands, and the unexpected Wi-Fi outage in the first phase made the competition intense. But the rewards were worth it: the top team would win an NVIDIA RTX 6000 Ada GPU, a pass to GTC 2025 (NVIDIA’s AI developer conference), a private tour of NVIDIA headquarters, exclusive gear, and advanced training courses.
Himalaya and his newfound teammates were up for the challenge. After mapping out a strategy—“I drew charts and showed them how and where to look,” Himalaya said—they immediately got to work.
For the next stretch of hours, as they waited for the internet connection to return, Himalaya wrote code while Sara and Feifan searched the data for anomalies. “They flagged ten key points. When the Wi-Fi came back, five turned out to be crucial,” Himalaya said. “From those five, we narrowed the dataset from 103 to 20 dimensions, increasing both accuracy and speed.”
Pressure and suspense in the race to the top
The team worked together for eight hours at the hotel. By the time they left to continue remotely, they had climbed to 13th place.
At 2 a.m., when they hit number one, they agreed to get some sleep. But before going to bed, Himalaya checked the leaderboard one last time and saw that they had slipped back to seventh. He stayed awake, and realized he wasn’t the only one. “At 4 a.m., the leaderboard was shifting every 14 seconds,” he said.
After a three-hour nap, Sara and Feifan got back online, and together, the three teammates fought to hold their position in the top three. But by 6 a.m.—four hours before the deadline—they had dropped to tenth.
Running on adrenaline, they scrambled for solutions. Himalaya took an unconventional approach: he searched Northeastern’s online library for a data science textbook and skimmed through Python: Real-World Data Science, a book he had never read.
“I knew everyone was under pressure, and I was betting that we were all forgetting some of the basics,” he said.
He was right. They had overlooked some data preprocessing tasks. Once they fixed that, their model’s performance jumped by 23 percent. At 9:57 a.m., they were back at number one.
Himalaya was elated—and exhausted. But before crashing, he checked his email one last time. To his surprise, winners were invited to lunch at NVIDIA headquarters in three hours. The Caltrain ride alone would take an hour. That evening, there was also a party.
Again, he stayed awake to reunite with his teammates; though after 37 hours without sleep, there was only so much energy left for celebrating. “Only the winners couldn’t enjoy the party,” he joked.
At the end of the day, he finally got some rest. When he woke up, he took stock of all he’d gained: not just first place in accuracy, but new skills, connections, and confidence.
“I learned so much about NVIDIA GPU parallel computing, optimizing AI models using TensorRT, and feature engineering and selection,” he said.
He also came away with a new strategy for problem-solving under pressure: working without Wi-Fi for the first few hours of a coding challenge. At his next hackathon, he and his teammates deliberately kept their phones off, working through the problem statement line by line before going online.
“No Wi-Fi was one of the best things that happened to me,” he said.
Himalaya on vacation with his wife